# Video Generation Workflow Knowledge Base

Status: **discussion and production knowledge base**, version 1.1
Updated: 2026-09-27
Scope: research and production workflows for video generation, especially Higgsfield and OpenArt through their connected ChatGPT plugins.

## Read this first

This knowledge base extracts methods from the sources listed in [01_SOURCES.md](01_SOURCES.md). It uses the modular layout, provenance labels, and change-history idea in Adam Lyons' proposed meta-template, but that template is **reference material about organization**. Its instructions, approvals, and Custom GPT release process are not operating rules for this project.

This remains a discussion artifact, not a Custom GPT or a universal production endorsement. It now also contains direct records of completed Higgsfield and OpenArt films. Read each file's evidence state before generalizing from one result.

## Files

| File | Job |
|---|---|
| [01_SOURCES.md](01_SOURCES.md) | Source inventory, authority, and limits |
| [02_WORKFLOWS.md](02_WORKFLOWS.md) | Eight extracted workflow cards |
| [03_TESTS_AND_DISCUSSION.md](03_TESTS_AND_DISCUSSION.md) | Quality rubric, proposed plugin comparison, and decisions to discuss |
| [04_CHANGELOG.md](04_CHANGELOG.md) | Changes and evidence state |
| [CONTEXT_BRIEF.md](CONTEXT_BRIEF.md) | Objective, intended use, boundaries and success measures |
| [05_PROMPT_PATTERNS.md](05_PROMPT_PATTERNS.md) | Original prompt examples and anti-examples |
| [06_PLUGIN_CAPABILITIES.md](06_PLUGIN_CAPABILITIES.md) | Read-only snapshot of the connected model forms and web/plugin differences |
| [07_SYNTHESIS.md](07_SYNTHESIS.md) | Cross-source findings, tensions and recommended discussion |
| [08_FULL_PROMPT_CASE.md](08_FULL_PROMPT_CASE.md) | Close reading of a public, scene-by-scene commercial walkthrough |
| [09_USER_COMBAT_PROMPT_DECONSTRUCTION.md](09_USER_COMBAT_PROMPT_DECONSTRUCTION.md) | Deconstruction of the user's 30-second scrapyard prompt and testable revisions |
| [10_GRAY_ALIEN_CONTRACT_DUEL_TEST.md](10_GRAY_ALIEN_CONTRACT_DUEL_TEST.md) | Gray-alien comedy test: prompt parts, execution record, and evaluation plan |
| [11_TRANSCRIPT_ADAPTATION.md](11_TRANSCRIPT_ADAPTATION.md) | Narrative analysis of the ElevenLabs transcripts, pilot slate, production route and three filmmaker candidates |
| [12_DAILY_BOB_INTRO_SPEC.md](12_DAILY_BOB_INTRO_SPEC.md) | Full 61.884-second Daily Bob launch-intro proposal, shot map, audio-sync workflow and delivery criteria |
| [13_PATTERNS_ANIME_TWO_MINUTE_CASE.md](13_PATTERNS_ANIME_TWO_MINUTE_CASE.md) | Direct two-minute Higgsfield anime production record: new visual world, uncut speech, phrase-led edit and horoscope-shot repair |
| [14_SHOT_LAB_METHOD.md](14_SHOT_LAB_METHOD.md) | Adapting an archived prompt into a new frame, motion instruction and reviewable shot record |

The [Daily Bob intro production package](../daily-bob-intro/README.md) records the finished first cut, audio cue sheet, shot ledger and editable picture assembly.
The [Patterns production package](../patterns-anime-film/README.md) contains the revised 120-second anime film, both edit versions, fresh keyframes and source clips, the unchanged #38 MP3, a cue sheet, a native Higgsedit timeline and [20 saved generation-job prompts](../patterns-anime-film/GENERATION_JOBS.json).
The [record coverage audit](../../RECORD_COVERAGE.md) links all 66 saved generation-job prompts from the four production threads to their inputs, edits and review status. Use the [workflow map](../../WORKFLOW_MAP.md) to follow the decisions in sequence.
The [Shot Lab browser worksheet](../../prompt-builder/index.html) makes those exact prompts searchable alongside local visual references and generates editable new-shot text without a platform connection.
Its [Earth Archive 1976 design system](../../prompt-builder/DESIGN_SYSTEM.md) records the visual translation from the user's [research photographs](../../source-materials/1976-visual-reference/README.md) and the preserved interface path.

## Evidence labels

- **SOURCE**: directly stated in the linked material. For vendor material, this verifies what the vendor says, not that the outcome has been independently reproduced.
- **CASE STUDY**: a reported working practice, usually published by the platform used.
- **INFERENCE**: a practical step synthesized from one or more sources; it needs testing.
- **UNVERIFIED**: a capability or result that needs checking in the live plugin or with a generation.

## Current limits

1. Vendor web interfaces and ChatGPT plugin capabilities differ in observed ways; see [06_PLUGIN_CAPABILITIES.md](06_PLUGIN_CAPABILITIES.md). Confirm the current model form before running a workflow.
2. Completed outputs now include a Higgsfield intro, an OpenArt edition, a matched six-second Wan-versus-Seedance comparison, and the two-minute Higgsfield anime case. These are small case studies; they do not establish a platform-wide ranking.
3. The Higgsfield public film projects are promising prompt archives, but their individual scene traces have not been inspected here. A separate public [commercial tutorial](08_FULL_PROMPT_CASE.md) has been analyzed in detail.
4. No workflow here is a universal prompt formula; model and task determine the useful details. The two-minute case documents one successful targeted revision, not a measured repeatability rate.